Realistic Models and Efficient Algorithms for Fault Tolerant Scheduling on Heterogeneous Platforms

Abstract

Most list scheduling heuristics rely on a simple platform model where communication contention is not taken into account. In addition, it is generally assumed that processors in the systems are completely safe. To schedule precedence graphs in a more realistic framework, we introduce an efficient fault tolerant scheduling algorithm that is both contentionaware and capable of supporting ε arbitrary fail-silent (fail-stop) processor failures. We focus on a bi-criteria approach, where we aim at minimizing the total execution time, or latency, given a fixed number of failures supported in the system. Our algorithm has a low time complexity, and drastically reduces the number of additional communications induced by the replication mechanism. Experimental results fully demonstrate the usefulness of the proposed algorithm, which leads to efficient execution schemes while guaranteeing a prescribed level of fault tolerance.

Iso-Level CAFT: How to Tackle the Combination of Communication Overhead Reduction and Fault Tolerance Scheduling

“…In other words, a given processor can simultaneously send a message, receive another message, and perform some computation. The bi-directional one-port model seems closer to the actual capabilities of modern networks (see the discussion of related work in [3,4]). Indeed, it seems to fit the performance of some current MPI implementations, which serialize asynchronous MPI sends as soon as message sizes exceed a few megabytes [3].…”

“…In this paper, we introduce the Iso-Level Contention-Aware Fault Tolerant (Iso-Level CAFT) scheduling algorithm (a new version of CAFT [3] that were initially designed to address both problems of network contention and fault-tolerance scheduling) that aims at tolerating multiple processor failures without sacrificing the latency. Iso-Level CAFT is based on an active replication scheme to mask failures, so that there is no need for detecting and handling such failures.…”

Fault tolerant heterogeneous MPSOC schedule length minimization based on platform reliability

“…(4) Note that if the task and its predecessor are scheduled on the same processor, there is no need to wait for other copies from task predecessor replicas to send their data to achieve fault tolerant operation since if the processor hasn't failed then task T will receive its data from the predecessor T` on the same processor, otherwise the processor is faulty and there is no need to send data to it [17].…”

“…CAFT (Contention-Aware Fault Tolerant) scheduling algorithm has been introduced in [17]. CAFT is another scheduling heuristic where tasks is replicated N times on N different processors to tolerate N-1 permanent processor failures however the number of communications induced by the replication scheme is drastically reduced.…”

“…The main idea is to schedule each task N times on N different processors so that in case of N-1 processor failures, there will be still a processor executing the task without errors thus tolerating N-1 processor error [17]. Active task replication has the advantage of making processor failures completely tolerable and transparent to the schedule since it doesn't require any extra action to detect or recover from failures.…”